[Paper Review] Acquiring Background Knowledge to Improve Moral Value Prediction

Ying Lin, Joe Hoover+3 more|arXiv (Cornell University)|Sep 16, 2017

Hate Speech and Cyberbullying Detection26 references3 citations

TL;DR

This paper proposes a context-aware framework that enhances moral value prediction in tweets by automatically acquiring and integrating background knowledge from external knowledge bases (e.g., Wikipedia) via entity linking. By enriching short, context-poor tweets with semantically relevant knowledge features, the model achieves performance comparable to a single human annotator, marking the first use of dynamic background knowledge for implicit psychological variable prediction in computational social science.

ABSTRACT

In this paper, we address the problem of detecting expressions of moral values in tweets using content analysis. This is a particularly challenging problem because moral values are often only implicitly signaled in language, and tweets contain little contextual information due to length constraints. To address these obstacles, we present a novel approach to automatically acquire background knowledge from an external knowledge base to enrich input texts and thus improve moral value prediction. By combining basic text features with background knowledge, our overall context-aware framework achieves performance comparable to a single human annotator. To the best of our knowledge, this is the first attempt to incorporate background knowledge for the prediction of implicit psychological variables in the area of computational social science.

Motivation & Objective

• To address the challenge of detecting implicit moral values in short, context-poor social media texts like tweets.
• To improve the accuracy of moral value prediction by incorporating external background knowledge beyond surface-level textual features.
• To explore whether knowledge acquisition from external sources such as Wikipedia can enhance NLP models for implicit psychological attribute detection.
• To demonstrate the feasibility of using entity linking and knowledge base enrichment as a generalizable method for computational social science applications.
• To provide a framework adaptable to other implicit sentiment or psychological variable prediction tasks.

Proposed method

• Entity linking is applied to identify named entities in tweets and link them to Wikipedia entries for knowledge acquisition.
• From linked entities, descriptive abstracts and properties are extracted and processed to identify terms highly correlated with each of the five moral foundations (Care/Harm, Fairness/Cheating, Loyalty/Betrayal, Authority/Subversion, Purity/Degradation).
• Correlated terms are used as additional discriminative features to augment standard textual features (e.g., n-grams, word embeddings) in a multi-label classification model.
• Part-of-speech tagging and dependency parsing are used to improve entity linking and feature extraction accuracy.
• The framework combines textual features with knowledge-derived features using a supervised learning model to predict moral value expressions.
• Manual selection of relevant entity and property types is used initially, with future work considering automated feature selection methods.

Experimental results

Research questions

• RQ1 Can background knowledge from external knowledge bases improve the detection of implicit moral values in short social media texts?
• RQ2 How effective is entity linking combined with knowledge base enrichment in enhancing moral value prediction compared to text-only models?
• RQ3 To what extent can knowledge about entities (e.g., Westboro Baptist Church) help resolve ambiguity in moral value inference?
• RQ4 What types of knowledge features (e.g., entity properties, descriptions) are most predictive of specific moral foundations?
• RQ5 How does the integration of background knowledge affect model performance relative to human annotators?

Key findings

• The proposed framework achieves performance comparable to a single human annotator in predicting moral values from tweets.
• Incorporating background knowledge significantly improves prediction accuracy over models relying solely on surface-level textual features.
• Entity linking enables the model to infer moral values associated with entities even when the moral reference is implicit, such as recognizing the moral implications of 'Westboro Baptist Church'.
• The integration of knowledge features derived from Wikipedia descriptions and properties enhances the model’s ability to detect subtle, context-dependent moral signals.
• Despite challenges, the method demonstrates that external knowledge can effectively compensate for the lack of contextual information in short texts like tweets.
• The study identifies limitations in entity disambiguation and knowledge relevance, particularly when knowledge bases contain static or overly general facts not aligned with the tweet’s context.